基于K-means的FITS图像自检验和自分类方法  

A self-verification and self-classification method for FITS images based on K-means

作　　者：曹晶 周浩[1] 戴智斌[2] 袁国武[1] CAO Jing;ZHOU Hao;DAI Zhibin;YUAN Guowu(School of Information Science and Engineering,Yunnan University,Kunming 650011,China;Yunnan Astronomical Observatory,Chinese Academy of Sciences,Kunming 650011,China)

机构地区：[1]云南大学信息学院,云南昆明650011 [2]中国科学院云南天文台,云南昆明650011

出　　处：《电子设计工程》2023年第23期180-183,195,共5页Electronic Design Engineering

基　　金：云南省重大科技专项(202202AD080004);国家自然科学基金项目(12263008);云南省应用基础研究计划重点项目(202001BB050032);中国自然科学基金重点项目(11933008);云南省“高层次人才培养支持计划”青年拔尖人才专项;云南省基础研究专项面上项目(202201AT070180);中国科学院西部之光面上项目。

摘　　要：针对如何快速实现地面观测获得的时序天文FITS(Flexible Image Transport System)图像的自动化分类和检验问题,提出了一种FITS图像自检验和自分类方法,该方法结合了K-means聚类算法及其思想,同时加入了一种基于皮尔逊相关系数的相似度算法。通过比较该方法与基于有监督的VGG13分类网络和基于无监督的K-means聚类算法应用于真实的天文数据分类得到的错误数量,得出该方法的分类准确率达94%以上。该方法一方面检验出了历史数据中存在的错误情况,摆脱了对关键词IMAGETYP和观测日志的依赖,进一步规范和完善了历史存储的天文FITS数据;另一方面增强了分类的可靠性,提高了数据获取效率,降低了人工成本。Aiming at the problem of fast automatic classification and verification of Flexible Image Transport System(FITS)images obtained from ground observation.A self-verification and self-classification method for FITS image types is proposed,which combines the unsupervised clustering method and its ideas,and adds a similarity algorithm based on the Pearson correlation coefficient.By comparing this method with the supervised VGG13 classification network and the unsupervised K-means clustering algorithm applied to real astronomical data to obtain the number of errors,it is concluded that the classification accuracy of this method is over 94%.On the one hand,it checks and corrects the error cases existing in the historical data,gets rid of the dependence on keywords IMAGETYP and observation logs,and further standardizes and improves the historical stored astronomical FITS data;on the other hand,it enhances the reliability of classification,improves the efficiency of data acquisition,and reduces the labor cost.

关 键 词：天文图像分类 K-MEANS聚类算法 皮尔逊相关系数 VGG13网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN99[自动化与计算机技术—计算机科学与技术]